Re: [libvirt] RFC: spec file cleanup ideas
On Fri, Apr 08, 2016 at 07:13:07PM -0400, Cole Robinson wrote:
Hi all,
I wanted to float some ideas about cleaning up the RPM spec file. None of it
is urgent but I'll end up scratching the itch eventually and I want to make
sure people are on board
Old distro bits:
* Drop support for building on rhel5... this was ACKd previously by eblake:
http://www.redhat.com/archives/libvir-list/2015-November/msg00175.html
Yep
* Drop support for end-of-life Fedora. There's lots of Fedora
conditionals
going back to Fedora 13. I don't think it's really interesting to try to
support building an RPM on a distribution that is no longer receiving updates.
But we could be a bit conservative and give a couple release window, dropping
support for less than Fedora 20 for example (which went EOL in June 2015)
There's a surprising/disappointing number of people who carry on running
EOL Fedora releases, so I think we should definitely be conservative in
this respect. 1 year grace after Fedora EOL seems sufficient though, so
dropping less than F20 would be OK.
Configuration variables: some of these could still feasibly be
'useful' even
if they aren't programmatically set by any distro conditional, but I think
unless someone actively uses them we should seek to have as few conditionals
as possible, rather than conditionalize everything by default which seems more
common. Some ideas:
Every conditional we have increases potential for bugs in the spec file
and I'm pretty sure we have some in there, because no one ever tests
most of these conditionals when disabled.
* Drop with_* conditionals that are only used by old Fedora or RHEL5.
I didn't
audit all of them but the obvious ones are straight forward like:
with_systemd_macros, with_polkit, with_capng, with_netcf, with_yajl,
with_capng, with_avahi, with_hal/all HAL support
For that matter I think we can drop HAL *code* entirely since it was
EOL after RHEL5, replaced by the udev driver.
* Drop client_only/server_drivers/with_libvirtd ... do these serve
any good
purpose? (once the rhel5 bits are gone)
The client stuff I'm a bit less sure about. The original rational for this
was that RHEL was only shipping libvirt with hypervisor drivers enabled on
certain architectures, and we wanted to enable other architectures todo
a client only build. eg to allow an ppc64 client to talk to a x86_64
host. Now that we have LXC enabled unconditionally on all archs though,
there's probably not a hugely compelling reason for a client only build
anymore.
* Drop with_driver_modules: it's unconditionally enabled and has
been since it
was introduced
Yep, that was there in case we needed to turn off drive modules for compatilbity
reasons on existing RHEL platforms. In the end we just switched RHEL over to
use them anyway, so they don't serve much use.
Old spec bits cleanup:
* drop defattr(), it's long since not needed
* drop manual removal of the buildroot, it's long since not needed
* use %global consistently instead of %define
* consider using %{with/without X} pattern instead of defining variables as with_X
Assuming %global is supported by RHEL6, its fine.
* The top of the spec has this block:
# If neither fedora nor rhel was defined, try to guess them from dist
%if !0%{?rhel} && !0%{?fedora}
%{expand:%(echo "%{?dist}" | \
sed -ne 's/^\.el\([0-9]\+\).*/%%define rhel \1/p')}
%{expand:%(echo "%{?dist}" | \
sed -ne 's/^\.fc\?\([0-9]\+\).*/%%define fedora \1/p')}
%endif
Is it actually still relevant? It was added 5 years ago... I don't know in
what case rhel or fedora macros won't be defined
In RHEL5 vintage build roots, %rhel/%fedora were not defined IIRC. No
longer an issue.
Regards,
Daniel
--
|: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org -o- http://virt-manager.org :|
|: http://autobuild.org -o- http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org -o- http://live.gnome.org/gtk-vnc :|